nn~ model attributes are model-defined, live-controllable parameters set via 'set NAME VALUE' messages

Beyond fixed methods, nn~ models expose attributes: model-specific parameters defined at export time. Attributes are set in the patch using a message box with the syntax set ATTRIBUTE_NAME VALUE. Examples include generation temperature, generation mode, or gain factor. The special built-in enable attribute bypasses computation without deleting the model. Attributes are model-dependent — there is no fixed list; you inspect them with nn.info or the dump message. This design allows live parameter control using any Max/Pd graphical object (number boxes, sliders, toggles) that sends the appropriate message.

Examples

Send set temperature 0.8 to control how random a generative model’s output is. Send set enable 0 to bypass the model with no CPU cost. Use a number box connected to a message set gain_factor $1 to sweep gain live.

Assessment

Describe two ways to discover what attributes a loaded model exposes. Write the Max message to set a float attribute named mix_level to 0.5. Explain what happens if you send set enable 0 vs. deleting the nn~ object.